PROMPT RADAR / VIDEO视频叙事
Claude Fable running ComfyUI workflows through the Comfy MCP is a cheat
Claude Fable running ComfyUI workflows through the Comfy MCP is a cheat code. → Claude pulled the shot from the web, auto-detected the scene cuts + trimmed it → ran my saved depthanything v3 + openpose workflow to build
@hellorob 3.7k 热度 2.8 万 浏览
编辑先看
它为什么值得进入灵感库
它最有价值的部分，是把一个模糊想法推进到了可以被模型执行的程度。 把一段描述拆成起始状态、动作变化、镜头响应和结束画面，连续性比形容词更重要。
01 / 原始方法
动作与镜头 Prompt
Claude Fable running ComfyUI workflows through the Comfy MCP is a cheat code. → Claude pulled the shot from the web, auto-detected the scene cuts + trimmed it → ran my saved depthanything v3 + openpose workflow to build the driving reference video → scrubbed every frame for the sharpest one, then character swapped me in with gpt-image-2 exactly how I directed it: wearing a tuxedo, liminal space environment, no background characters → wrote the seedance 2.0 prompts + ran the gens → stitched together the 3-panel comparison you're watching The preprocessing + prompt writing is usually most of the work on a clip like this. With the MCP I didn't touch any of it. Didn't need to look at a single node lol.
02 / 复刻路径
不要一次改完所有变量
用 GPT Image 2 复刻时，建议先单独生成稳定首帧，再把动作拆成“开始、变化、收束”三段；每段只保留一个主要运动。
- 主体：Claude Fable running ComfyUI workflows through the Comfy MCP is a cheat code. → Claude pulled the shot from the web, auto-detected the scene cuts + trimmed it → ran my saved depthanything v3 + openpose workflow to build
- 视觉方向：实时 AI 视频案例；查看原帖媒体与作者说明
- 镜头与构图：按原帖镜头流程执行；查看原帖媒体与工作流
- 交付约束：auto，保留原作者方法与归属；复刻时避免水印、低清、畸变、主体漂移和无关文字。
负向约束
保留原作者方法与归属；复刻时避免水印、低清、畸变、主体漂移和无关文字。
模型建议
GPT Image 2 / Seedance / Veo / Kling。先用原始 Prompt 建立基准结果，再根据模型对自然语言、镜头运动或文字排版的能力做局部改写。
来源与编辑原则